Rescue of collapsed replication forks is dependent on NSMCE2 to prevent mitotic DNA damage

NSMCE2 is an E3 SUMO ligase and a subunit of the SMC5/6 complex that associates with the replication fork and protects against genomic instability. Here, we study the fate of collapsed replication forks generated by prolonged hydroxyurea treatment in human NSMCE2-deficient cells. Double strand breaks accumulate during rescue by converging forks in normal cells but not in NSMCE2-deficient cells. Un-rescued forks persist into mitosis, leading to increased mitotic DNA damage. Excess RAD51 accumulates and persists at collapsed forks in NSMCE2-deficient cells, possibly due to lack of BLM recruitment to stalled forks. Despite failure of BLM to accumulate at stalled forks, NSMCE2-deficient cells exhibit lower levels of hydroxyurea-induced sister chromatid exchange. In cells deficient in both NSMCE2 and BLM, hydroxyurea-induced double strand breaks and sister chromatid exchange resembled levels found in NSCME2-deficient cells. We conclude that the rescue of collapsed forks by converging forks is dependent on NSMCE2.

Introduction Replication-associated DNA damage is common in human cells and can lead to the development of somatic mutations. DNA damage during replication can be induced by DNA lesionproducing chemicals, proteins bound to the DNA, DNA polymerase inhibitors, or nucleotide limitation [1]. Hydroxyurea (HU) triggers fork stalling due to nucleotide limitation through inhibition of ribonucleotide reductase [2]. Human cells exposed to HU for up to six hours are capable of restarting 80% of their replication forks [3][4][5][6]. However, forks that are stalled for 16 to 24 h are unable to restart [7], indicating they have collapsed. Collapsed replication forks must be rescued by active forks initiated at dormant origins to complete genome duplication. In both human cells and yeasts, the induction of a double strand break (DSB) is associated with repair of collapsed forks [8,9]. Although factors essential for the formation of DSBs during fork collapse have been identified [10], the mechanism generating DSBs when the new replication forks converge with the collapsed forks is unknown.
The DNA helicase mutated in Bloom's syndrome BLM possesses multiple functions in DNA replication fork stabilization and homologous recombination (HR), which is a mechanism that operates in the repair of replication-associated DSBs [11]. Recruitment of BLM to replication forks is part of the mechanism that stabilizes forks in both unperturbed and replication-stressed cells [12,13]. Excessive DSBs accumulate in BLM-deficient cells released from replication blockade after prolonged fork stalling [14], suggesting that BLM plays a role in collapsed-fork rescue. In the absence of BLM, after fork collapse, under-replicated DNA and unresolved HR intermediates persist into mitosis where they cause DNA damage [15,16].
The E3 SUMO ligase NSMCE2 is a component of the SMC5/6 complex, which is present at stalled replication forks and a key component of the stalled fork proteome [17]. In budding yeast, deletion of the NSMCE2 homolog MMS21 is lethal; however, sumoylation-deficient hypomorphs are viable and have defects in replication-specific DNA repair [18]. These cells also accumulate excess RAD51-dependent recombination intermediates during replication stress and are deficient in HR [18,19]. During fork stalling, MMS21 undergoes auto-sumoylation during replication stress, and it then recruits the BLM homolog Sgs1 via SUMO binding sites on Sgs1 [20,21]. Once recruited, Sgs1 resolves HR intermediates generated during repair of damaged replication forks.
In human cells, forks adopt a RAD51-dependent structure during stalling, which resembles a Holliday junction [22]. RAD51 is required to prevent replication-induced DSBs, and RAD51 levels increase at stalled forks as they transition from a restart-competent state to a collapsed state [7]. BLM regulates the exchange of RAD51 recombinase for RPA [23,24], and in previous work we showed that sumoylation of BLM regulates a switch between BLM's pro-and anti-recombinogenic functions [14]. If negative regulators of RAD51 such as BLM and the recently described RADX are ablated, excess RAD51 is loaded at stalled forks and excess DSBs accumulate [25]. In other situations, however, induction of excessive RAD51 can instead trigger inhibition of HR repair [26]. Because NSMCE2 regulates BLM recruitment and RAD51-dependent HR intermediates accumulate in yeast mms21 mutant cells, we hypothesized that NSMCE2 may be a critical regulator of RAD51 function at collapsed replication forks.
Here we studied the role of NSMCE2 in repair and rescue of collapsed replication forks. We found that NSMCE2 is essential for formation of DSBs during collapsed-fork rescue. Interestingly, lack of DSBs during collapsed-fork rescue is associated with hyper-accumulation of RAD51 and impaired sister chromatid recombination. Defects in the rescue of collapsed replication forks in NSMCE2-deficient cells lead to DNA damage in mitosis.

Results
We used two different siRNAs to deplete NSMCE2 in HeLa cells, resulting in an approximately 80% reduction in both RNA and protein levels (S1A Fig). To corroborate key NSMCE2-deficient phenotypes, we also prepared NSMCE2-/-clones of HEK293T cells in which we targeted a single site in exon 2 of NSMCE2 and isolated two clones with different frameshift mutations containing no detectable NSMCE2 protein (S1B Fig). Hereafter, we refer to cells in which we have knocked down NSMCE2 as NSMCE2-deficient cells and we refer to NSCME2-/-cells as NSMCE2 null cells.

NSMCE2 is required for BLM sumoylation and its localization to stalled replication forks
Experiments in yeast suggested that NSMCE2 is required for efficient sumoylation of BLM [20]. To measure SUMO-BLM levels, we used human U2OS cells that express a His-tagged SUMO2 to carry out pull down assays. Analysis of SUMO-conjugates revealed that sumoylated BLM levels increased approximately eight fold after prolonged fork stalling by treatment with 2 mM HU for 16 hours (Fig 1A and 1B). We then tested if sumoylation of BLM is dependent on NSMCE2 by knockdown of endogenous NSMCE2. Depletion of NSMCE2 using two different siRNAs resulted in a 60% decrease in sumoylated BLM in HU-treated cells compared to HU-treated control cells (Fig 1A and 1B). These data indicate that BLM sumoylation is dependent on NSMCE2. Residual SUMO-BLM could result from the incomplete depletion of NSMCE2 or from residual sumoylation catalyzed by other E3 SUMO ligases.
We previously reported that a BLM protein mutated at its preferred sumoylation sites K317R and K331R is recruited normally to stalled replication forks [14]; consequently, we hypothesized that BLM localization at stalled forks might be normal in NSMCE2-deficient cells. On the contrary, siRNA-mediated depletion in HeLa cells led to three-fold reduction in BLM foci in cells treated with HU for 24 h compared to HU-treated control cells ( Fig 1C). These data show that efficient recruitment to or retention of BLM at collapsed replication forks is dependent on NSMCE2.

NSMCE2 prevents excessive accumulation of RAD51 and promotes HR
BLM can promote dissociation of RAD51 recombinase from ssDNA [23]. Because BLM's role in dissociation of RAD51 at stalled forks could be defective in NSMCE2-deficient cells, we tested whether NSMCE2 plays a role in regulation of RAD51 accumulation. NSMCE2-deficient cells, both untreated cells and cells treated with HU for 24 hours, exhibited increases in the number, intensity, and size of RAD51 foci compared to controls (Fig 1D and 1E [7,27], we tested whether the excess RAD51 was localized to collapsed replication forks. To test this, we labeled nascent DNA synthesis with 10 μM EdU for 12 min prior to HU treatment, treated cells with HU for 24 hours, and then analyzed RAD51 foci. As expected, RAD51 co-localized with EdU foci in HU-treated control and NSMCE2-deficient cells (S3C Fig). These data show that NSMCE2 is required to prevent over-accumulation of RAD51 at forks under conditions that lead to their collapse.
RAD51 is normally loaded onto ssDNA by exchange with ssDNA binding protein RPA [28]. We therefore tested whether the excess RAD51 accumulation in NSMCE2-deficient cells might correlate with a diminished accumulation of RPA at stalled forks. For this experiment, we measured the accumulation of chromatin-bound RPA after nucleoplasmic extraction of cells treated with HU for 24 hours. After siRNA-mediated depletion of NSMCE2 and HU treatment, cells displayed 40% fewer RPA foci in both HeLa (Fig 1F and 1G We conclude that RAD51 accumulates in excess over RPA in NSMCE2-deficient and NSMCE2 null cells, perhaps due to a failure to recruit BLM to stalled forks. The lower levels of RPA foci suggest that there are lower levels of ssDNA. To test this possibility, we measured the levels of ssDNA by incorporation of BrdU for two cell divisions prior to HU treatment, followed by immunodetection with anti-BrdU antibodies in non-denaturing conditions. Unlike control cells, which displayed at least a two-fold increase in BrdU foci after treatment with 2 mM HU for 24 hours, NSMCE2-deficient cells displayed no induction of BrdU foci after HU treatment (S3H Fig; representative low power images of BrdU localization are shown in S4D Fig). Thus, the lower levels of focal RPA in HU-treated NSMCE2-deficient cells are evidence of lower levels of ssDNA detectable by anti-BrdU antibodies at collapsed replication forks. Because anti-BrdU antibodies cannot detect BrdU in the ssDNA-RAD51 nucleoprotein filament [29], these results do not rule out the possibility that the excess RAD51 is bound to ssDNA.
To test whether the RAD51-bound chromatin in NSMCE2-deficient cells is competent for HR, we measured the frequency of sister chromatid exchanges (SCEs) after prolonged fork stalling by 24-hour treatment with HU. The SCE assay measures crossovers generated after resumption of DNA synthesis that can be detected in the subsequent mitosis. NSMCE2-deficient cells had a 45% reduction in the number of HU-induced SCEs/metaphase compared to control cells (Fig 1H). Thus, the excess RAD51 observed at stalled forks is not associated with increased sister chromatid recombination. Contrary to previous reports using murine cells [30], we found that basal levels of SCEs in HEK293T NSMCE2 null cells were similar to normal HEK293T cells (

NSMCE2 is required for DSBs and RAD51 resolution during collapsedfork rescue
Because HU-induced SCEs were suppressed in NSMCE2-deficient cells, we hypothesized that the excess RAD51 leads to a defect in DSB formation during rescue of collapsed forks. To test this possibility, we measured DSB accumulation in control and NSMCE2-deficient cells after prolonged exposure to HU. In HeLa cells exposed to control siRNA, a 16-hour treatment with HU did not induce DSBs; however, a 48-hour treatment with HU led to an accumulation of DSBs detectable by pulsed-field gel electrophoresis (PFGE) (Fig 2A). Interestingly, we found that NSMCE2-deficient cells failed to produce a detectable increase in DSBs after a 48-h exposure. Ionizing radiation with 4 Gy followed by a 30-min repair period results in equal levels of DSBs in both control and NSMCE2-deficient cells, indicating that NSMCE2-deficient cells are not defective in DSB formation per se but in replication stress-induced DSBs.
Because NSMCE2-deficient cells are defective in DSB formation at stalled forks after prolonged HU treatment, we tested whether the DNA damage response was also reduced. We measured γ-H2AX levels by analysis of DNA damage foci and flow cytometry. We found by both measures that NSMCE2-deficient cells accumulate two-to three-fold less γ-H2AX after prolonged HU treatment (S3J and S3K Fig To investigate the ability of cells to generate DSBs during collapsed-fork rescue, we measured the kinetics of accumulation of DSBs over time after release from HU. After release from the HU block, normal cells linearly accumulated DSBs, whereas NSMCE2-deficient cells failed to accumulate DSBs four and eight hours after release (Fig 2A). The accumulation appears to be replication-dependent, because normal cells released into 10 μM aphidicolin after HU arrest did not accumulate similar levels of DSBs (S5A Fig). Flow cytometry confirmed that control and NSMCE2-deficient cells show similar cell-cycle distributions 6 and 12 hours after release from HU, suggesting that differences in cell-cycle progression after release from HU do not explain these results (S5B Fig). Moreover, no significant differences in the levels of apoptosis were observed in control and NSMCE2-deficient cells after release from HU, ruling out apoptosis as a confounder of differences in DSBs (S5D Fig). Levels of DSBs in untreated HEK293T NSMCE2 null cells were higher than in untreated normal HEK293T cells; however, similar to the results obtained with NSMCE2 depletion with siRNAs, after treatment with HU and during release into normal medium we observed a defect in accumulation of DSBs in NSMCE2 null cells (S5C Fig). Similar to NSMCE2-deficient HeLa cells, NSMCE2 null cells were also defective in their γ-H2AX response after HU treatment (S5E Fig). Collectively, the results suggest that in the absence of NSMCE2 the levels of DSBs that are normally generated during collapsed-fork rescue are reduced.
We next tested whether NSMCE2-deficient cells have a defect in the dynamics of RAD51 localization during collapsed-fork rescue. Because RAD51 protein accumulates during HU treatment, we hypothesized that converging forks displace the RAD51 over time. We therefore released cells from prolonged fork stalling and measured levels of the RAD51 foci at collapsed replication forks in a time course. Two, four, and eight hours after release from HU, control HeLa cells exhibited a steady decrease in RAD51 foci whereas NSMCE2-deficient cells retained them ( Fig 2B). RAD51 foci increased in normal cells between 8 and 12 h after release from HU, possibly due to RAD51-dependent DNA repair in late S or G2 phase. In addition, we also observed a persistence of RAD51 foci at stalled forks in NSMCE2 null cells compared to normal cells after release from HU treatment (S6A Fig). In both normal and NSMCE2 null HEK293T cells, RAD51 foci co-localized with RPA and γ-H2AX foci (S6B Fig). We considered the possibility that persistence of excessive RAD51 at collapsed replication forks might disturb replication dynamics in S phase after release from HU. To measure replication fork dynamics, we performed microfluidics-assisted replication track analysis (maRTA) [31]. We found that replication fork speed, fork restart, and dormant origin firing were similar in NSMCE2-deficient cells compared to control cells, after either 5 or 16 hours of HU treatment ( Fig 2C). These data indicate that both the replication dynamics of unperturbed forks and of dormant origin activation in replication-stressed cells are not adversely affected by NSMCE2 deficiency.

NSMCE2 prevents mitotic DNA damage resulting from replication stress
NSMCE2-deficient cells maintained normal cell-cycle progression in the absence of HU treatment ( Fig 2D); however, after release from prolonged HU treatment, NSMCE2-deficient cells displayed an arrest in the next G1 phase (S7A Fig). Defects in collapsed-fork rescue could lead to under-replicated DNA in S phase and DNA damage in mitosis. Similar to previously reported results [30,32], we found increases in the frequencies of abnormal anaphases, micronuclei, and G1 53BP1 nuclear bodies after release from HU block in NSMCE2-deficient cells compared to controls indicating that excess mitotic DNA damage is induced in NSMCE2-deficient cells ( To investigate further the nature of the mitotic damage invoked in HU-treated, NSMCE2deficient cells, we measured the frequency of ultra-fine bridge (UFB) formation in cells undergoing mitosis. In order to obtain a sufficient number of cells in anaphase, cells were pretreated or not with HU for 24 hours; they were then blocked in G2 with the CDK1 inhibitor RO-3306 at 7.5 μM for 15 hours and then finally released into metaphase for 1 hour prior to fixation ( Fig 3A). Flow cytometry confirmed effective G2 arrest by RO-3306 treatment (S7G Fig). We visualized UFBs using the PICH repair helicase, which localizes to UFBs and DNA under tension [3]. The number of UFBs measured by PICH staining was increased after HU treatment in NSMCE2-deficient cells but not control cells (Fig 3B-3D and 3F). PICH-positive UFBs were also positive for BLM (Fig 3C), indicating that localization of BLM to these structures is not dependent on NSMCE2. The crosslink repair protein FANCD2 is sometimes associated with the ends of UFBs, and has been used as a marker for under-replicated DNA persisting into mitosis [3,16]. FANCD2-flanked, PICH-positive UFBs ( Fig 3D) were infrequently observed in both NSMCE2-deficient and control cells. Thus, the excess UFBs produced in NSMCE2-deficient cells are not equivalent to the UFBs produced in cells treated with lowdose aphidicolin [15,16]. We then tested whether mitotic DNA damage originated from collapsed forks in the previous S phase. As a marker for damaged DNA and repair in anaphase cells, we counted the number of FANCD2 foci and found a 1.8 fold increase in NSMCE2-deficient cells treated with HU compared to control cells (Fig 3E and 3G). In order to monitor the location of collapsed forks generated by prolonged treatment with HU, we labeled cells with EdU for 20 min before treatment with HU. FANCD2 foci co-localized with the EdU label at a greater frequency in HUtreated, NSMCE2-deficient cells compared to control cells (Fig 3H), indicating that the recruitment of FANCD2 observed in mitosis had occurred in regions of chromatin where replication forks had previously stalled and collapsed. These data suggest that defective collapsed-fork rescue in NSMCE2-deficient cells leads to increased under-replicated DNA persisting into mitosis, which results in mitotic DNA damage.
In order to measure DNA damage in metaphase cells, we measured the levels of γ-H2AX associated with metaphase chromosomes and the levels of chromosome aberrations detectable at metaphase. For the analysis of γ-H2AX levels, normal HEK293T and NSMCE2 null cells were treated with 2 mM HU for 24 hours, released into medium with RO-3306 to block them in G2, then released into normal medium and prepared for analysis of γ-H2AX levels by immunofluorescence (S8A and S8B Fig). HU-treated NSMCE2 null cells exhibited a nearly 50% increase in median fluorescence intensity of chromosome-associated γ-H2AX in phospho-H3-positive cells compared to HU-treated HEK293T normal cells. For the analysis of chromosome aberrations, cells were treated or not with 2 mM HU for 24 hours and then metaphase chromosomes were prepared and analyzed by fluorescence microscopy (S9A Fig). We identified increased frequencies of chromatid arm breaks, telomere fusions, and secondary constrictions in NSMCE2 null cells compared to control cells. Chromatid arm breaks and secondary constrictions were induced by HU treatment (S9B Fig). The increase in secondary constrictions in HU-treated NSMCE2 nulls cells is consistent with increased under-replicated DNA and the increase in chromatid breaks could arise from chromosome breakage in mitosis or under-replication as seen at common fragile sites.

Epistasis relationships among NSMCE2, BLM, and RAD51 during collapsed-fork rescue
Because NSMCE2 is essential for proper localization of BLM to stalled replication forks but displays phenotypes distinct from BLM-deficient cells, we asked whether NSMCE2 is epistatic to BLM during rescue of collapsed forks. In these experiments, we used siRNAs to deplete BLM in NSMCE2 null and control cells (Fig 4A). In HU-treated normal HEK293T cells depleted for BLM, levels of phosphorylated RPA and γ-H2AX were similar to levels in HUtreated control cells as evidence by Western blot analysis. In contrast, in HU-treated NSMCE2 nulls cells depleted for BLM, the levels of phosphorylated RPA and γ-H2AX were reduced in comparison to HU-treated control cells, but they were similar to levels in HU-treated NSCME2 null cells. The data for phosphorylated H2AX were confirmed by analysis of focal γ-H2AX and flow cytometry with γ-H2AX antibodies (Fig 4B, 4D and 4E). Prolonged HU treatment of BLM-deficient normal HEK293T cells resulted in a 2.6 fold increase in the levels of SCEs from 8.2 to 21.1 SCEs/metaphase, whereas prolonged HU treatment of BLM-deficient NSMCE2 null cells resulted in an only small increase in SCEs from 4.6 to 6.3 SCEs/metaphase ( Fig 4C). Consistent with the suppression of HU-induced SCEs, the levels of HU-induced DSBs were also suppressed in BLM-depleted NSMCE2 null cells relative to BLM-deficient normal cells. These data indicate that NSMCE2 is epistatic to BLM with respect to HU-induced phenotypes.
Because DSB accumulation is suppressed in NSMC2-deficient cells, we tested whether depletion of RAD51 would restore HU-induced DSB levels to normal in NSMCE2 null cells. After transfection of control and RAD51 siRNAs, we treated cells with 2 mM HU for 24 hours, then released into normal medium for 6 hours and quantitated the DSB marker γ-H2AX by flow cytometry and measured DSBs by PFGE. In HU-treated, control-depleted normal HEK293T cells, γ-H2AX levels increased approximately 10 fold compared to baseline after release into normal medium for 6 hours (S10A and S10C Fig). Contrary to expectation, RAD51 depletion in normal cells resulted in much smaller induction of γ-H2AX, similar to the levels observed in HU-treated control-depleted NSMCE2 null and HU-treated, RAD51-depleted NSMCE2 null cells. Consistent with the γ-H2AX results, in HU-treated, controldepleted normal HEK293T cells, DSB levels increased approximately three fold compared to baseline after release into normal medium for 6 hours (S10B Fig). In contrast, RAD51 depletion in normal cells resulted in almost no induction of DSBs compared to baseline, which again was similar to the levels of DSBs in HU-treated control-depleted NSMCE2 null and HUtreated, RAD51-depleted NSMCE2 null cells. These data suggest that, despite the fact that RAD51 has hyper-accumulated at collapsed replication forks, in the absence of NSMCE2, the RAD51 at collapsed forks is nonfunctional.

Discussion
DNA combing experiments in many mammalian cell lines have shown that, after prolonged fork stalling due to HU exposure, DNA synthesis does not normally resume at the site where the fork stalled (see reference 2 and Fig 2). These data rule out rescue mechanisms, such as HR-mediated restart or break-induced replication, in which replication is re-established at the site of fork stalling. The majority of collapsed forks must therefore be rescued by converging forks initiated at dormant origins after release from prolonged replication arrest. DSBs have been previously associated with resumption of DNA synthesis after release from prolonged HU block [7,14], but the mechanism by which these breaks are generated is not well understood. Here we show that normal rescue of collapsed replication forks is dependent on NSMCE2. After resumption of DNA synthesis, NSMCE2-deficient cells do not accumulate normal numbers of DSBs. The large increase in UFBs in mitosis indicates that many forks fail to complete DNA replication.
NSMCE2 deficiency is associated with higher levels of RAD51 foci at collapsed forks and persistence of foci after release from replication arrest. These results are consistent with results obtained in yeast mutants of the SMC5/6 complex and of MMS21, in which excess RAD51-dependent recombination intermediates accumulate at stalled forks [19,33,34]. Pathological accumulations of RAD51 have been associated with DNA repair defects [26,35]. We do not know the structure of the RAD51-bound DNA in NSMCE2-deficient cells, and the excess RAD51 could be at the fork itself, associated with ssDNA gaps behind the fork, or associated with other abnormal structures. There are established roles for RAD51 at stalled replication forks that do not involve DSBs. For instance, RAD51's role in reversal of the replication fork is epistatic to its BRCA2-mediated fork protection function [36,37]. Our experiments with RAD51 depletion in normal cells suggest that reversed fork structures, catalyzed by RAD51, are required for the formation of DSBs during fork rescue, very likely as substrates for nucleases such as MUS81. Biochemical studies suggest that BLM performs a quality-control function on stressed replication forks by dissociating nonfunctional, ADP-bound RAD51 from the nucleoprotein filament [23]. Because BLM does not accumulate normally at stalled forks in the absence of NSMCE2, it is possible that the RAD51 that accumulates excessively in NSMCE2defcient cells is in the nonfunctional ADP-bound state. That said, the epistasis experiments indicate that NSMCE2 controls other factors besides BLM that contribute to the function of RAD51 at collapsed forks. We propose a model in which the excess nonfunctional RAD51 prevents DSB formation during the convergence of active replication forks with collapsed forks, leading to excess under-replicated DNA that is detectable at anaphase (Fig 5). However, more Black line indicates separate blots using lysates from the same experiment. The experiment was performed two times. (B) Analysis of focal concentrations of γ-H2AX after treatment with HU for 24 hours and after release into normal medium for 6 hours. Box and whisker plots representing at least 10,000 γH2AX foci in two independent experiments. Medians are shown above the graph. (C) Levels of HUinduced SCEs. Box and whisker plots representing distribution of SCE per metaphase. Medians of the combined data are shown above the graph. Two independent experiments were performed. Unpaired t-test was performed to analyze the distributions. (D) Analysis by flow cytometry of fluorescence intensity of γ-H2AX after treatment with HU for 24 hours and after release into normal medium for 6 hours. Median florescence intensity was measured from a minimum of 10,000 events. The bar graph represents the fold change in median fluorescent intensity normalized to untreated normal HEK293T cells exposed to control siRNA from a single experiment. Two independent experiments were performed. (E) Representative distributions of fluorescence intensity of cells in selected conditions from the flow cytometry data. (F) Analysis by PFGE of DSBs after treatment with HU for 24 hours and after release into medium for 6 hours. The bar graph represents fold change in DSBs detected by PFGE normalized to untreated normal NSMCE2 cells exposed to control siRNA. The error bars represent the SEM of three independent experiments. The gel below the bar graph contains the results from one experiment. See Fig 1 for definition  NSMCE2 and collapsed-fork rescue definitive tests of this proposition are required to rule out other possible models. For example, it could be informative to use molecular combing to monitor the replication dynamics of forks from newly fired origins as they converge upon collapsed replication forks.
We found that BLM sumoylation is dependent on the presence of NSMCE2. We and others [32] have shown that BLM does not accumulate normally at stalled forks in NSMCE2-deficient cells. BLM has multiple functions in the resolution of recombination intermediates during replication stress, normally ensuring that recombination intermediates are resolved without exchange. Yet the levels of SCEs are low in the absence of NSMCE2, suggesting additional roles of NSMCE2 in promotion of crossover events when BLM is absent. Our evidence suggests that RAD51-depenedent intermediates in NSMCE2-deficient cells are not resolved until M phase, whereas RAD51 foci are normally resolved during S phase. Because BLM localization to UFBs is not dependent on NSCME2, BLM could have a role in resolving RAD51-depenedent intermediates in mitosis. Our findings that excess γ-H2AX accumulated on metaphase chromosomes but not during S phase suggest that under-replicated DNA is not resolved until G2/M phase. For example, prometaphase DNA repair [3] or mitotic resolvases [38] could decatenate under-replicated DNA to permit disjunction of inter-linked sister chromatids.
We cannot rule out the possibility that the loss of NSMCE2 in cells affects the function of the SMC5/6 complex. However, in agreement with previous results in human U2OS and DT40 cells [39,40], we found SMC5 levels were normal in NSMCE2-deficient HeLa cells (S2B Fig). These data suggest that, unlike in S. cerevisiae [41], SMC5 levels are not dependent on the presence of NSMCE2 in human cells. Whether human NSMCE2 plays a structural role in collapsed-fork rescue and other repair processes remains to be determined. Normally (top part of model), after exposure to HU, uncoupling of the replicative helicase from the replicative DNA polymerase at the fork leads to production of ssDNA, which is immediately bound by ssDNA binding protein RPA (red circle). BLM (yellow circle) is one of the first factors to be recruited to the stalled replication fork and this accumulation of BLM is dependent on the NSMCE2 protein (green rectangle) on the SMC5/6 complex (purple ring). The localization of BLM at the stalled fork assists in maintaining functional RAD51 (teal rhombus) exchange during fork reversal. Given enough time in the reversed configuration, the fork is unable to restart (fork collapse). After resumption of DNA synthesis at dormant origins, an active fork converges upon the collapsed fork. The convergence is associated with RAD51 dissociation and a DSB at the collapsed fork, followed by DSB repair by HR via the canonical RAD51-mediated pathway. In NSMCE2-deficient cells (bottom part of model), BLM is no longer recruited to the stalled fork, and there is a failure in the regulation of RAD51 loading at the fork, shown here as a loss of dynamic exchange of RAD51 and RPA resulting in excess, hyper-stable RAD51. After resumption of DNA synthesis at dormant origins, the aberrant RAD51 structure prevents convergence of the active fork with the collapsed fork, resulting in a double fork stall instead of a DSB. The double fork stall persists into mitosis where it results in an ultra-fine DNA bridge and mitotic DNA damage. https://doi.org/10.1371/journal.pgen.1007942.g005

NSMCE2 and collapsed-fork rescue
Our results here agree with previous results showing lower levels of SCEs in NSMCE2-deficient cells [42]. Hypomorphic NSMCE2 mutation in humans is associated with a syndrome characterized by short stature and acanthosis nigricans [32]. Cells derived from these patients display increased micronuclei, nuceloplasmic bridges at cytokinesis, and binucleated cells. Despite a defect in BLM localization at replication forks in patient cells, untreated cells have normal SCE levels, and UV treatment induces only a small increase in SCEs [32]. We found that cells deficient for both NSMCE2 and BLM exhibit reduced levels of HU-induced DSBs and SCEs, indicating that NSMCE2 is epistatic to BLM during collapsed-fork rescue. In contrast to results with human cells, murine cells that are null for Nsmce2 exhibit increased SCEs, and Blm knockdowns in the murine Nsmce2 null cells display an additive increase in SCE levels [30]. The explanation for these different outcomes of NSMCE2 deficiency between humans and mice is unknown.
We used HU to generate and study collapsed forks and to block repair-coupled DNA synthesis (e.g., break-induced replication, gap filling, lesion bypass, etc.). We therefore cannot rule out the possibility that NSMCE2 plays other roles during the unperturbed cell cycle or in situations where template switching can bypass DNA damage during replication, such as in cells treated with methyl methanesulfonate or UV irradiation. We observed higher levels of DSBs in untreated NSMCE2 null cells (S4 Fig), but observed no increase in basal SCE levels ( Fig 4C). Our analysis using maRTA indicated that deficiency of NSMCE2 did not alter replication dynamics in untreated cells, which is consistent with results reported in yeast [43]. We suggest that the increased DSBs in untreated NSMCE2-deficient cells may originate from incomplete replication at common fragile sites, leading to DNA breakage in mitosis and the formation of micronuclei in the next cell cycle. Micronuclei are known to be prone to replication-associated DNA breakage [44]. HR-directed DSB repair is dependent on NSMCE2 [45,46]; consequently, these breaks would most likely be repaired by non-homologous end joining. Studies in both mammalian cells [30,47] and yeasts [48] indicate that HR can be increased in NSMCE2-deficient cells under some conditions, emphasizing the general complexity of NSCME2's roles in maintenance of genomic integrity.
The importance of mechanisms that regulate RAD51 protein levels is underscored by studies that have identified increased RAD51 protein levels as a negative predictor of patient outcome in several cancer types [49]. The present work has uncovered a connection between NSMCE2 and the formation of DSBs at collapsed replication forks during rescue. The identification of NSMCE2 as a potential controller of HR-mediated fork rescue highlights NSMCE2's potential as a new therapeutic target for combinatorial therapy of HR-dependent cancers.

Genome editing for NSMCE2
Genome editing was carried out with Integrated DNA Technologies (IDT) ALT-R CRISPR--Cas9 system with crRNA guide (GTCCATACCAGAGTTGATAC) targeting the first coding exon. Ribonucleoprotein particles were introduced into HEK293T cells using FuGENE HD (Promega). After 48 hours, single cells were deposited in a 96-well plate by flow cytometry. After four to six weeks, clones were analyzed by the T7 endonuclease assay (New England Biolabs), and clones that scored positively were PCR sequenced. >50 clones were analyzed in the first screen and a single heterozygous NSMCE2 null mutant was obtained. This clone was genome edited again to obtain two NSMCE2 null mutants.

SUMO2 pull-down
U2OS cells that were stably transfected with a His-tagged SUMO2 were kindly provided by Dr. Michael Matunis at Johns Hopkins University, who obtained them from Dr. Mary Dasso's lab at NIH. The cells were reverse-transfected with siRNAs using LifeTechnologies' Lipofectamine RNAiMAX. Cells were treated with 2 mM HU for 16 hours. His-SUMO2 conjugates were purified as described by Tatham et al. [51]. Briefly, cells were washed with ice-cold phosphate-buffered saline (PBS) and directly lysed in 4 ml lysis buffer (6 M guanidine-HCl, 100 mM NaCl, 10 mM Tris-HCl pH 7.4, 3 mM imidazole and 2 mM β-mercaptoethanol) and sonicated to reduce viscosity. Lysates were incubated with 50 ml Talon Metal Affinity Resin (Clontech Laboratories, Inc.) overnight at 4˚C with gentle mixing, and then washed 2 times with 4 ml guanidine wash buffer (6 M guanidine-HCl, 100 mM NaCl, 10 mM Tris-HCl pH 7.4, 0.1% Triton X-100, 2 mM β-mercaptoethanol) followed by 3 washes in 4 ml urea wash buffer (8 M urea, 100 mM NaCl, 10 mM Tris-HCl pH 7.4, 0.1% Triton X-100, 2 mM β-mercaptoethanol). The beads were transferred to 1.5 ml microfuge tubes for one more wash with urea wash buffer and proteins were eluted for 1 hour at room temperature with elution buffer (63 mM Tris-HCl pH 6.8, 2% SDS, 200 mM imidazole, 1.5% β-mercaptoethanol, 10% glycerol, with bromophenol blue). The eluates were boiled for 5 min and cleared by centrifugation prior to loading on a 10% polyacrylamide gel. The whole cell lysates were lysed in RIPA buffer. Laemmli buffer was added to 1X in aliquots representing 10% of the eluate, then boiled for 5 min prior to gel loading.

Analysis of chromatin-bound RPA
The protocol from Mendez and Stillman for chromatin isolation by small-scale fractionation was followed [52]. HEK293T normal and NSMCE2 null cells treated or not with 2 mM HU for

Immunofluorescence and image analysis
The protocol was adapted from Dimitrova and Gilbert [53]. Cells were grown on coverslips overnight and washed with cold CSK buffer (10 mM HEPES pH 7.4, 300 mM sucrose, 100 mM NaCl, 3 mM MgCl 2 ) and nucleoplasm was extracted for 90 seconds with cold extraction buffer (0.5% Triton X-100 in CSK buffer with 1 mM PMSF, 50 mM sodium fluoride, 0.1 mM sodium orthovanadate and 1x EDTA free Halt protease inhibitor) prior to 30-min fixation in 4% formaldehyde at room temperature. Cells were washed twice with cold PBS then treated with 0.5% Triton X-100 at room temperature before staining. Cells were blocked at room temperature for 1 hour using sterile filtered 3% BSA in PBS, then probed using primary antibodies for 1 hour at room temperature. Secondary antibodies (Alexa Fluor 488 and 546) were used at 1:1000 for 45 min and nuclei were stained using Molecular Probes NucBlue reagent (R37606). For 5-ethynyl-2´-deoxyuridine (EdU) labeling, cells were incubated in 10 μM EdU for 20 min. EdU labeling and detection was performed using Life technologies Click-iT EdU Alexa Fluor 647 imaging kit according to manufacturer's instructions. Cells were mounted in Molecular Probes ProLong Gold Antifade Reagent. Fixed and stained cells were imaged using the Leica SP5-II spectral confocal microscope using the 63x/1.4 NA PL Apo objective. Using the nuclear signal to mask the region of interest enabled accurate measurement of the number and intensity of nuclear foci and the percent of nucleus occupied by signal for each antibody target. No less than 10,000 foci were analyzed per experimental group. Box and whisker plots were used to visualize the distribution of foci. For analysis of EdU labeled forks and ultra-fine bridges (UFBs), z-stacks were created using 100x objective and deconvolved using the GE DeltaVision Elite High Resolution Microscope. Images were analyzed and 3D representations were created using the NIS Elements software.

Flow cytometry
Cells were harvested with trypsin/EDTA, resuspended in ice cold PBS, and fixed and stained using BioLegend True-Nuclear Transcription Factor Buffer. Cells were then stained for γ-H2AX using directly conjugated antibody and counterstained using 7AAD to monitor cell DNA content. For cell-cycle assays, analysis was performed using the BD Pharmingen FITC BrdU Flow Kit. A minimum of 30,000 events was recorded for each group using the BD FACS-Canto II or BD LSR II flow cytometer. Apoptosis analysis was carried out using the BioLegend FITC Annexin V Apoptosis Detection Kit with propidium iodide (PI).

Cytogenetic analysis
For SCE analyses, cells were cultured with 10 μM BrdU (Sigma-Aldrich). After 60 hours, the cells were incubated with 0.02 μg per ml colcemid (Invitrogen) for up to 2 hours, harvested and processed as described earlier [14]. For the epistasis experiments using HEK293T cells, the cells were incubated in 0.6 μg per ml colcemid for 16 hours prior to harvest. The slides were examined under the microscope at 100×, and SCEs were counted from metaphases with an acceptable quality of sister-chromatid discrimination. For measurements of HU-induced SCEs, cells were cultured in 10 μM BrdU for 30 hours, washed one time with 1× PBS, and treated with 2 mM HU for 24 hours. Cells were then released into medium containing 10 μM BrdU for an additional 20 hours. Metaphases were collected in colcemid and processed as described above. For analysis of micronuclei, normal or NSCME2-depleted HeLa cells were seeded onto coverslips, treated or not with 2 mM HU for 24 hours, and fixed with 4% formaldehyde. The cells were stained with Nuc-Blue (Thermo Fisher) at 2 drops per ml in PBS for 30 min, washed briefly, mounted, and imaged using the GE DeltaVision Elite High Resolution Micro.
For analysis of chromosomes at metaphase, normal and NSMCE2 null HEK293T cells were seeded into 60 mm dishes and incubated overnight. For analysis of γ-H2AX labeled chromosomes, cells were treated with 2 mM HU for 24 hours, then washed and incubated in medium with 7.5 μM RO-3306 for 10 hours for HEK293T control or 20 hours for NSMCE2 null cells. Cells were then released into normal medium and harvested at the indicated time points. Cells were fixed with 4% formaldehyde and stained with anti-phospho-histone H3 (serine 10). For analysis of chromosome spreads, cells were treated with 0.02 μg per ml colcemid for 45 min or with 0.6 μg per ml colcemid for 16 hours prior to harvest. Cells from both experiments were harvested and metaphases prepared as described [14]. Metaphase chromosomes were stained with Nuc-Blue. Cells or chromosomes were imaged using the GE DeltaVision Elite High-Resolution Microscope.

DSB detection by pulsed-field gel electrophoresis
The procedure was performed as previously described [8,14]. Sub-confluent cultures of HeLa cells or HEK293T cells untreated, treated with 2 mM HU for 24 hours, or treated with HU for 24 hours and released into normal medium for different times were harvested by trypsinization. Agarose plugs of 2.5 x 10 5 cells were prepared in disposable plug molds (Bio-Rad Laboratories). In some experiments, cells were released into medium containing 10 μM aphidicolin for 24 hours. Plugs were then incubated in lysis buffer (100 mM EDTA, 1% wt/vol sodium lauroyl sarcosinate, 0.2% wt/vol sodium deoxycholate, and 1 mg/ml proteinase K) at 37˚C for 16 hours. Plugs were then washed four times in 20 mM Tris-HCl, pH 8.0, and 50 mM EDTA before loading onto an agarose gel. Electrophoresis was performed for 21 hours at 14˚C in 0.9% (wt/vol) agarose containing Tris-borate/EDTA buffer in a PFGE apparatus (CHEF DR III; Bio-Rad Laboratories), according to the following protocol: block I: 9 hours, 120˚included angle, 5.5 V/cm, 30 to 18-s switch; block II: 6 hours, 117˚included angle, 4.5 V/cm, 18 to 9-s switch; block III: 6 hours, 112˚included angle, 4.0 V/cm, 9 to 5-s switch. The gel was then stained with SYBR Gold (1 part in 10,000 in water; Invitrogen) and analyzed by the AlphaImager system (ProteinSimple). Relative DSB levels were assessed by comparing DSB signals for each treatment to the background levels observed in untreated conditions using Image J. Data were analyzed with GraphPad Prism software.

Microfluidic-assisted replication track analysis
maRTA was performed as previously described with some modifications [31]. Briefly, 36 hours after siRNA transfection, HeLa cells were pulse-labeled with 50 μM iododeoxyuridine (IdU) for 40 min. Cells were then treated or not with 2 mM HU for 5 hours or 16 hours. The cells were released in fresh medium containing 50 μM of chlorodeoxyuridine (CldU) for 40 min. Cells were then harvested and embedded into agarose plugs containing 20,000 cells/plug. After proteinase K digestion and agarose digestion by beta-agarase, DNA fibers were stretched on 3-aminopropyltriethoxysilane coated slides (LabScientific) using polydimethylsiloxane molds fashioned with micro-capillary channels prepared as described [31]. DNA fibers were then denatured in 2.5 M HCl, and probed with the following antibodies: mouse IgG 1 anti-BrdU/IdU (clone BD44, Becton Dickinson), rat anti-BrdU/CldU (clone B1/75, Bio-Rad OBT0030), and mouse IgG 2a anti-ssDNA (clone [16][17][18][19]Millipore). Secondary antibodies included Alexa Fluor 488 anti-mouse IgG 1 , Alexa Fluor 594 anti-rat, and Alexa Fluor 647 antimouse IgG 2a , respectively (Life Technologies). Images were acquired on Leica DMI6000 epifluorescence microscope using Leica LAS-AF software. Signals were measured using NIH Ima-geJ software with custom-made modifications and the data analyzed with GraphPad Prism software.

Statistics
Statistical evaluations of experiments with continuous variables (e.g., quantitation of SUMO-BLM and DSBs) or discrete variables in which the data was normal were carried out by paired Students t-test. Evaluations of experiments with non-normal discrete variables (e.g., focal counts) were analyzed by Mann-Whitney test. All p-values were two-sided. . HeLa cells were exposed to control or NSMCE2 siRNAs and treated or not with 2 mM HU for 24 hours, then released into fresh media for 16 hours before fixation and staining. The graph plots percent values of normal mitoses (blue), mitoses with anaphase bridges (red), and mitoses with lagging chromosomes (green). The value shown above the graph is the percent of abnormal mitosis scored. Representative images of normal mitosis (B'), mitosis with an anaphase bridge (red caret in B"), and mitosis with a lagging chromosome (red caret in B"') are shown (right panel). At least 100 anaphase cells were analyzed in each experimental condition. Three independent experiments were performed. (C) Quantitative analysis of micronuclei formation after release from HU block (left panel). Experiment was performed three times and analyzed by chi square test. A representative image is shown to the right of the graph. Scale bars represent 10 microns. (D) Quantitation of 53BP1 nuclear bodies/cell using bins of 1, 2, and 3 or more bodies/cell. HeLa cells were exposed to control or NSMCE2 siRNAs and treated with 2 mM HU for 24 hours, then released into fresh media for 24 hours before fixation and staining. Each bar represents the mean percent of total cells observed in each class and the error bars represent the SEM values. Flow cytometric analysis showed that the majority of cells are in G1 (see panel A). Cells with large nuclei indicative of being in the G2 phase were excluded from the quantitation. Three independent experiments were performed. (E) Quantitative analysis of area of the nucleus inhabited by 53BP1 signal after release from HU block based on the same data in panel D. Box and whisker plot represents area distribution per cell in at least 250 cells in each experimental condition. (F) Representative images of cells stained for 53BP1 analyzed in panels D and E. (G) Flow cytometric analysis of the effect of the treatment schedule to enrich for mitotic cells used in the experiments shown in Fig 3. HeLa cells were transfected with control or NSMCE2 siRNAs. After 24 hours, they were treated or not with 2 mM HU for 24 hours followed by release into medium containing 7.5 μM RO-3306 for 15 hours, followed by release into normal media for 1 hour. Cells were harvested, fixed, and stained with 7-AAD. 10,000 events were analyzed in each experimental condition. This experiment was performed two times.  Fig. RAD51 and NSMCE2 are epistatic with respect to HU-induced phenotypes. Analysis of HU-induced phenotypes in normal HEK293T and NSMCE2 null cells in which RAD51 levels were reduced or not by siRNA-mediated depletion. (A) Analysis by flow cytometry of the fluorescence intensity of γ-H2AX after treatment with HU for 24 hours followed by release into normal medium for 6 hours (Rel). The error bars represent the SD of median fluorescence intensity from a minimum of 10,000 events in five independent experiments. (B) Analysis by PFGE of DSBs after treatment with HU for 24 hours followed by release into medium for 6 hours (Rel). The bar graph represents the mean fold change in DSBs detected by PFGE normalized to untreated normal HEK293T cells exposed to control siRNA. The error bars represent the SD of three independent experiments. The inset gel shows the results from one experiment.